Reliability of Recent Global Digital Elevation Models for Geomatics Applications in Egypt and Saudi Arabia

Global Digital Elevation Models (DEMs) have been utilized in various geomatics activities worldwide. Recently, there exist several available DEMs vary significantly in terms of spatial resolution and release dates. This paper examines the reliability of eight recent global DEMs, namely the EarthEnv-D90, SRTM 1, SRTM 3, ASTER, GMTED2010, GLOBE, GTOPO30, and AW3D30, in two study areas in Egypt and Saudi Arabia representing different topography patterns. Known ground control points with measured accurate coordinates and precise elevations have been utilized in evaluating the performance of those DEMs. It has been concluded that such a judgment procedure should not be carried based on a single statistical measure. First, five statistical measures, specifically the range, standard deviation, correlation, kurtosis, and skewness, have been evaluated separately for each DEM’s errors. Then, a new reliability index is introduced based on the weighted average concept. The accomplished results show that global DEMs perform differently in different topography patterns. It has been concluded that the EarthEnv-D90 and SRTM1 models attain high reliability indexes in the Nile delta region that represents a flat topography, while the GMTED2010 and EarthEnv-DEM90 models came in the first places for the second study area, Makkah, which represents mountainous topography.

Although there are many available free global DEMs, their accuracy and reliability should be quantified prior to actual implementation [13] [14] [15] [16]. This paper investigates the accuracy and reliability of eight existing global DEMs based on several statistical measures. Table 1

Data and Study Areas
Two study areas have been utilized in this research: the Nile delta region, Egypt and Makkah metropolitan area, Kingdom of Saudi Arabia (KSA). These two test areas have been selected so that the first one represents flat topography environment, while the second area is representing a hilly or complex topography region. The first study area covers most of the Nile delta region north of Egypt from Alexandria at the west (29.6˚E) to Port Said at the east (32.3˚E) on the Mediterranean coasts (Figure 1(a)). It is bounded by the two branches of the Nile River, and extending a little bit from both sides to cover coastal areas with existing ground data, with an overall area of approximately 13,000 square kilometers.
The topography of the region smoothly slopes northwards towards the Mediterranean coast, where the difference in elevation between its southern peak, at Cairo, and northern coastal fringes is approximately 18 meter. 416 control points with measured GPS coordinates along with precise elevations (relative to the Mean Sea Level: MSL datum) have been utilized in this research study. Those points lay mainly on the two Nile branches and on the coasts of the Mediterranean, with an average spacing of 5 kilometers (Figure 1(a)).
The second study area is located at south west of KSA, about 80 kilometers east of the Red sea and extending from latitude 21˚09'N to 21˚37'N and from longitude 39˚35'E to 40˚02'E, covering 1301 square kilometers approximately ( Figure   1(b)). Its elevations vary from almost 100 meters in the west to more than 960 meters in the west. It is a matter of reality that the topography of Makkah is complex in nature, and several mountainous areas exist inside its metropolitan area. Within this region, 137 ground control points with precise MSL-based elevations and GPS coordinates are known and utilized in the research study.

Methodology and Data Processing
For global DEM accuracy assessment, the measured elevation of each control point (H m ) has been compared against its corresponding DEM-based elevation (H DEM ), and the height differences or DEM errors (∆H) is then estimated: The outlier detection procedure took place before any further analysis in order to detect and remove erroneous observations or gross errors. An outlier is, simply, an observation that appears to depart significantly from the remaining observation in the sample. There are many statistical approaches for outlier detection, but the traditional method is the Z score statistical test as: Journal of Geographic Information System where, Y i is an observation, Y − denotes the sample mean, and SD is the sample standards deviation. The computed Z i score, for each observation, is compared against the values obtained from normal distribution table. If the score exceeds the given limits, it is considered as a possible outlier and should be removed.
Next, five statistical measures have been evaluated individually [17], and then a unique reliability index has been proposed and computed. The correlation between observed elevations of the control points (X or H M ) and their respective elevation from a particular DEM (Y or H DEM ) is computed as: where COR represents the correlation between X and Y, are the summation of X, Y, and XY respectively, and n is the total number of available sample points.
Subsequently, the range of height difference is evaluated as the difference between the maximum and minimum computed differences for each DEM. Thirdly, the standard deviation (SD) of height differences is computed as: where n is the number of the utilized control points. It is expected that the larger the standard deviation, the flatter the normal distribution curve. The smaller the standard deviation, the higher the peak of the curve. Several research studies [18] [19] depend primly on the standard deviation measure in evaluating DEMs performance, which is not the case in this study. Afterward, a histogram of each DEM errors is plotted, and the skewness and kurtosis statistical measures are evaluated. The skewness is computed as: and, The kurtosis, as another measure of the shape of the histogram or the normal distribution curve, is computed as: Having those five statistical measures attained for each DEM, they are categorized in a certain manner and a rank is assigned to each DEM in each measure (R i ). For the correlation, range, and standard deviation measures, the ranks vary where W ∑ denotes the sum of the utilized weight, i.e. 10.

Results and Discussion
The eight selected DEMs have been downloaded, from their respective websites, for both study areas and the elevation of each control has been interpolated using the ARC GIS 10 software package. The statistics of the attained results are presented in Table 2. It can be realized from that table that the AW3D30 and EARTHEnv-DEMM90 have the biggest correlation to the observed values in both study areas respectively. Next, the outlier detection procedure has been applied using the Z-score approach. Table 3 Table 4, and their histograms are plotted in Figure 4(a) and Figure 4 Table 4 for both study areas.
The last procedure computes the reliability index of each DEM in each study in each study areas as a weighted mean (Equation (14)) based on the ranks of the models in the five statistical measures, and the accomplished results are tabularized in Table 5(a) and

Conclusions
This research study has investigated the reliability of eight recent global digital elevation models for geomatics and GIS applications. Those models include EarthEnv-D90, SRTM 1, SRTM 3, ASTER, GMTED2010, GLOBE, GTOPO30, and AW3D30. The evaluation has been carried out over two study areas: the Nile delta region, Egypt, representing a flat topography, and Makkah metropolitan area, Saudi Arabia representing a hilly or complex topography region. 416 and 137 control points with measured GPS coordinates along with precise elevations have been utilized in both areas respectively. First, the outlier detection procedure has been applied using the Z-score approach in order to increase the accuracy and integrity of the datasets. It has been concluded that the improvements, or the reduction of the standard deviation, have averages of 8.6% and 10.1% for both study areas respectively. Accordingly, the outlier detection should be a regular routine applied in similar studies.
Based on the accomplished results, it has been concluded that such a judgment procedure should not be carried based on a single statistical measure. A novel re-Journal of Geographic Information System liability index is introduced based on the weighted average concept and has been applied in two study areas in both Egypt and Saudi Arabia. The attained results showed that the EarthEnv-D90 and SRTM1 models attain high reliability indexes in the Nile delta region that represents a flat topography. In addition, for the second study area, Makkah, which represents mountainous topography, the GMTED 2010 and EarthEnv-DEM90 models came in the first places on the relative reliability scale of ten. Moreover, the accomplished findings reveal that global DEMs perform better in flat area than in hilly or complex topography. Furthermore, it is concluded that even though the accuracy measures of global DEMs are not high, they present a good candidate when accurate local models are not available.